Professional Documents
Culture Documents
= 80
to 1120. Mean and second-order statistics downstream of Re
U
ref
y
L
U
cyl
(a)
x
U
ref
=1
y
L =1
U
cyl
= 0.7 U
ref
(b)
0
Figure 2. (a) Layout in the experiments of Liu & Rodi (1991); (b) layout in the present numerical
simulation; the computational domain is dened as 0.1 6 x/L 6 3.5, 0 6 y/L 6 0.8, 0 6 z/L 6 0.2.
1992; Henningson, Lundbladh & Johansson 1993) and in ows experiencing oblique
transition (Berlin, Lundbladh & Henningson 1994).
Previous research on bypass transition induced by free-stream turbulence left open
to interpretation the issue of whether intensication and ultimate destruction of the
streaky structures arises from boundary layer internal dynamics, or from forcing by
free-stream eddies. Such ambiguity is primarily due to experimental diculties in
following the details of the generation and growth of disturbances because of their
randomness in space and time.
Figure 3(ac) illustrates three scenarios observed in the present investigation (see
, 4). In the most interesting case of those drawn in gure 3, inlet wake disturbances
rapidly evolve into pus similar to those found in Westin et al. (1994) and Alfredsson
& Matsubara (1996) and turbulent spots appear downstream. More usually, the pus
decay as in gure 3(b), at least below the critical Reynolds number Re
= 200.
Turbulent eddies inside the passing free-stream wake impinge on the boundary layer
and sometimes interact with its outer part, such as to subject the ow to a rapidly
growing instability (, 4). This involves an intensication of the near-wall streaky
structures, and eventually the breakdown into young turbulent spots. Figures 3(d)
and 3(e) depict two other scenarios in which the ow is subjected to very strong
or extremely weak disturbances from the passing wake. Observations concerning the
Boundary layer transition induced by passing wakes 113
(a)
Puff
Forcing
Breakdown
(young spot)
Turbulent
spot
Turbulent
strip
Further decay Decay
(b)
(c)
Forcing
Young spot Decay
(d)
Turbulent patch
(e)
Elongated structure with spanwise modulation
z
x
Figure 3. Several possibilities for the downstream propagation of certain types of inlet distur-
bances. The sketches represent u
/
in an (x, z)-plane very close to the wall: (a) intermediate-strength
disturbance and strong forcing; (b) intermediate-strength disturbance and weak forcing; (c) inter-
mediate-strength disturbance with downstream strong forcing; (d) strong disturbance; (e) very weak
disturbance.
last two instances have been made in the present study through additional numerical
simulations and thought experiments (see , 7).
1.2. The connection with some engineering turbomachinery research
Experiments on wake-induced periodic unsteady transition in turbomachinery blade
rows were reported by Dring et al. (1982), Dong & Cumpsty (1990), Addison &
Hodson (1990), Mayle & Dullenkopf (1991), and Halstead et al. (1997) (see also
114 X. Wu, R. G. Jacobs, J. C. R. Hunt and P. A. Durbin
the reviews by Mayle 1991 and Walker 1993). Halstead et al. (1996) reported mea-
surements in compressors and low-pressure turbines. Their experiments showed that
transition in unsteady, turbomachine boundary layers develops along two dierent,
but coupled, paths. These consist of a wake-induced strip under the convecting wake
trajectory, and a path between wakes that is caused by other disturbances. Along
both paths the boundary layer goes from laminar to transitional to turbulent, with
large regions of laminar and transitional ow. The switch from the non-wake path
to the wake-induced strip was found to occur in a small fraction of a blade passing
period. Halstead et al. noted that assumptions of predominantly turbulent boundary
layers on multi-stage turbomachine blading are incorrect.
Although turbine and compressor experiments have indicated that the eects of
wake passing can be substantial and have provided useful guidelines for further
research, the technical complexities involved in obtaining detailed quantitative data
from rotating turbine/compressor stages make it dicult to isolate physical mecha-
nisms. Realizing such complexities, a number of investigators have considered simpler
geometries. In the simplest of these (Pfeil, Herbst & Schroder 1983; Liu & Rodi 1991;
Orth 1993; Zhong et al. 1998) the unsteady blade row interaction was simulated by
sweeping a row of wake-generating cylinders past a at plate (gure 2). Liu & Rodi
(1991) obtained time- and phase-averaged mean and uctuating streamwise velocity
proles for four dierent wake passing frequencies. In their experiments, the Reynolds
number was fairly low so that the boundary layer remained laminar over the full plate
length when no disturbing wakes were present. They found that the wake-produced
turbulent strips grew together and caused the boundary layer to become fully turbu-
lent. The streamwise location of the merger moved upstream with increasing wake
passing frequency.
Using experimental data gathered from a similar ow conguration, Orth (1993)
concluded that in turbomachinery ows periodically disturbed by passing wakes,
the disturbance enters the boundary layer very early on, and convects within it
before leading to transition. Periodic uctuation in the velocity prole, as opposed
to stochastic uctuation, does not have a major inuence on the transition. This is
consistent with the present study. Orth (1993) also suggested that the location where
transition takes place is only dependent on inlet turbulence intensity: the passing
wake exerts no eect on the process. Our study shows that this needs qualication.
Inception of turbulent spots in wake-induced transitional ows is intimately linked
with turbulent eddies of the travelling free-stream wake. One additional pleasant
connection of the present study with turbomachinery research concerns the recent
liquid crystal visualization experiments of Zhong et al. (1998) and Kittichaikarn et
al. (1999). This work was communicated to us by Prof. Hodson. Our turbulent spots,
as well as their embryo precursors, resemble those observed by Zhong et al. (1998)
and Kittichaikarn et al. (1999) to a remarkable degree (see , 3 and , 4).
2. Mathematical and numerical considerations
2.1. Problem denition
Consider the evolution of an incompressible ow over a smooth at plate with
upstream wakes passing periodically (gure 2b). The origin of the coordinate system
is at the leading edge of the plate. The wakes are assumed to be generated by
imaginary circular cylinders positioned in the plane x = L and moving in the y-
direction at U
cyl
, which can be either positive or negative, corresponding to inlet wakes
Boundary layer transition induced by passing wakes 115
traversing away from, or towards, the at plate. The velocity of the ow upstream
of the cylinder is U
ref
. The cylinders are equally spaced so that they cut through
the y = 0 plane at a specied passing period . The characteristic velocity scale is
U
ref
, the characteristic length scale is L, the Reynolds number is then Re = U
ref
L/
where is the kinematic viscosity of the uid. Throughout this study Re = 1.5 10
5
,
as in Liu & Rodi (1991). The mean ow properties of the wake are determined by
free-stream velocity, cylinder velocity and cylinder diameter.
The computational domain for the DNS is dened as 0.1 6 x/L 6 3.5, 0 6 y/L 6
0.8, and 0 6 z/L 6 0.2. The inlet momentum thickness Reynolds number is Re
= 80
in all the simulations. Unless otherwise noted, all velocities are normalized by the
reference velocity U
ref
and all lengths by the characteristic length scale L.
2.2. Governing equations and notation
Mass and momentum conservation is enforced for ow over the at plate by solving
the full time-dependent, mass-conservation and NavierStokes equations in Cartesian
coordinates,
div u = 0, (1)
u
t
+ div (uu) =
1
grad p + div
_
1
Re
_
grad u + (grad u)
_
, (2)
where u is the velocity vector with Cartesian components (u, v, w) or u
i
, i = 1, 2, 3.
Superscript denotes transpose. The equations are in non-dimensional form.
In this paper, time-averaging is represented by . Averaging at a particular phase,
t
m
n
= m+ n
6 1 is the
fraction of the wake passing period. For example, the phase-averaged mean velocity
components are evaluated as
(u
i
)(t
n
) =
1
M
M
m=1
u
i
(t
m
n
), (3)
where M is the total number of periods within which phase averaging is per-
formed. Averaging over the homogeneous spanwise z-direction is implied in both
time-averaging and phase-averaging. Time-averaged and phase-averaged mean veloc-
ities are related via u
i
= (u
i
). Thus the instantaneous velocity can be decomposed
as
u
i
= (u
i
)(t
n
) + u
/
i
(t
n
) = u
i
+ u
i
(t
n
) + u
/
i
(t
n
), (4)
where u
i
(t
n
) = (u
i
)(t
n
) u
i
is the periodic velocity uctuation with respect to
the time-averaged mean, and u
/
i
(t
n
)
_
u
j
(u
j
)(t
n
)
_
dn
. (5)
2.3. Inow and other boundary conditions
For all the simulations described in this paper, the height of the computational
domain 0.8L is approximately 200 at the inlet x = 0.1L, 20 in the middle of the
plate x = 1.75L, and 11 at the exit x = 3.5L. (x) is the 99% boundary layer
thickness. The width of the computational domain 0.2L is equivalent to 40 at the
inlet, 5 in the middle of the plate and 3 at the exit.
116 X. Wu, R. G. Jacobs, J. C. R. Hunt and P. A. Durbin
Depending upon the passing frequency 1/, at any given instant either one single
wake or multiple wakes can be found on the inow plane. A wake will be found
on the computational inow plane if its centreline is located within the range from
y = b cos to y = 0.8L+b cos , where b is wake half-width and = tan
1
U
cyl
/U
ref
is the wake inclination angle (gure 2). The total number of such wakes at any time
is
.= CEILING
_
0.8L + 2b cos
U
cyl
_
, (6)
where the function CEILING returns the smallest integer greater than or equal to its
real argument.
Consider U
cyl
< 0. On the inow plane at the beginning of each wake passing
period (t
m
n
= 0 the wake starts to enter the top boundary of the inow plane. At any
subsequent time 0 < t
m
n
U
cyl
. Each point on the inow plane then has an
eective coordinate with respect to the origin of the wake-coordinate system, which
is y
e, wake
(y) = (y y
centreline
)/cos .
For U
cyl
> 0, the origin of the wake-coordinate system starts to move upwards from
y = b cos . This indicates that at t
m
n
U
cyl
. Each point on the inow plane then has an
eective y
wake
-coordinate with respect to the origin of the wake-coordinate system,
which is y
e, wake
(y) = (y y
centreline
)/cos .
At the inow station of the computational domain x = 0.1L, the velocity compo-
nents were prescribed as
u = u
blasius
+ u
blasius
_
cos
.
q=1
u
e, wake, q
sin
.
q=1
v
e, wake, q
_
,
v = v
blasius
+ u
blasius
_
sin
.
q=1
u
e, wake, q
+ cos
.
q=1
v
e, wake, q
_
,
w =
.
q=1
w
e, wake, q
,
_
_
(7)
where subscript blasius denotes the steady Blasius prole. The subscript e, wake, q
represents the eective instantaneous velocity components in the qth wake-coordinate
system. The velocities (u
e, wake, q
, v
e, wake, q
, w
e, wake, q
) are the same as the wake velocities
(u
wake
, v
wake
, w
wake
) illustrated in gure 2 if y
e, wake, q
6 b, and zero otherwise. The
wake velocities were multiplied by u
blasius
so they would satisfy no slip where the wake
intersects the plate. As the wake enters the computational domain, a more realistic
prole develops rapidly. Distortion of the velocities at the inlet results in a localized
streamwise pressure gradient. The eect of such a pressure gradient on transition and
boundary layer development will be studied in , 7 through numerical experiments.
In general, the mean velocity prole in wakes at a large distance from a solid
body is independent of the shape of the body, except for a scale factor (Schlichting
1979). Liu & Rodi (1991) also pointed out that near the leading edge of the plate
the cylinder wake is self-similar, and has also lost its shedding characteristics. Raj &
Lakshminarayna (1973) demonstrated that the wake behind an airfoil trailing edge is
Boundary layer transition induced by passing wakes 117
similar to that of a two-dimensional cylinder wake. Invoking self-similarity allows the
wake introduced at the inlet to be obtained more simply than by actually computing
a real cylinder wake.
The turbulent wake velocities (u
wake
, v
wake
, w
wake
) appearing in (7) were generated
from a separate precomputation of a temporally decaying, self-similar plane wake,
following the work of Moser, Rogers & Ewing (1998) and Ghosal & Rogers (1997) .
In a temporally decaying wake, the ow is statistically homogeneous in the streamwise
and spanwise directions, and inhomogeneous in the cross-stream direction. The initial
conditions for the temporally decaying plane wake simulation were generated from a
turbulent channel ow simulation at Reynolds number 3300 based on the centreline
mean velocity and channel half-height. The procedure involved taking two realizations
of half-channel ow and fusing them together. Physically this corresponds to a
situation in which two half-channel ows exist on either side of a rigid plate and
the plate is instantaneously removed. This simulation was performed on a grid size
of (65, 128, 128) using an LES code (Wu & Squires 1997). The grid sizes used by
Ghosal & Rogers (1997) and Moser et al. (1998) were (65, 48, 16) and (512, 195, 128),
respectively.
Mean ow and turbulence statistics of the simulated plane wake are presented in
gure 4. All velocities in the gure are normalized by the maximum mean velocity
decit u
wake, max
; lengths are normalized by the wake half-width b. Following Ghosal
& Rogers (1997) , the half-width b is dened as the distance between the two points at
which the mean velocity decit is 50%u
wake, max
. This is slightly larger than the distance
between the wake centreline and the rst point with eectively zero mean velocity
decit. Figure 4(a) shows that mean velocity proles obtained at the three dierent
instants (indicated in the caption by their descending maximum velocity decits)
collapse. These mean proles are also in excellent agreement with experimental
measurements and the data correlation of Schlichting (1979), i.e. u
wake
/u
wake, max
=
[1 (y
wake
/1.1338b)
1.5
]
2
. This demonstrates that the simulated wake has reached a
self-similar state and that the mean ow has lost its memory of the initial condition.
At these and subsequent instants, the product of the wake width and maximum decit
remains constant.
Figure 4(b, c) show the r.m.s. turbulence intensities for the decaying plane wake
at the same three instants. Turbulent shear stress proles are given in gure 4(d).
The turbulence intensities obtained by Moser et al. (1998) and Ghosal & Rogers
(1997) have two distinct features: double peaks occur in the streamwise component,
and the wall-normal component is slightly higher than the spanwise component.
In addition, the results of Ghosal & Rogers show that unlike the mean ow and
the anti-symmetrical Reynolds shear stress proles, the turbulence intensities are
not self-similar as time proceeds. It is evident from gure 4(bd) that the present
precomputation reproduced all these essential features. Figure 4(e) presents proles
of the rate of turbulence kinetic energy dissipation. Spanwise energy spectra of the
turbulence kinetic energy are given in gure 4(f) for completeness. Note that the
results in gure 4 are presented in the wake-coordinate system (see gure 2). The
uctuating wake velocities obtained from the precomputation are rescaled by the
wake maximum decit u
wake, max
and half-width b before they are applied to (7).
Using the experimental correlation of Schlichting (1979) for far wakes, at x/L = 0.1
u
wake, max
= 0.14U
ref
and b 0.1L for the present ow conditions.
At the top surface of the computational domain the following boundary conditions
were applied: v = v
blasius
, u/y = v/x, and w/y = v/z. This is articial, but
given the substantial distance between the top surface and the wall, the eect of the
118 X. Wu, R. G. Jacobs, J. C. R. Hunt and P. A. Durbin
(e)
0.20
0.16
0.12
0.08
0.04
0
2 1 0 1 2
y/b
b
/
u
3w
a
k
e
,
m
a
x
( f )
10
10
1 10 100 1000
k
z
b
(c)
0.8
0.6
0.4
0.2
0
2 1 0 1 2
y/b
(d)
0.10
0.06
0.02
0.02
0.06
0.1
2 1 0 1 2
y/b
(a)
0.1
0.1
0.3
0.5
0.7
1.1
2 1 0 1 2
y/b
(b)
0.8
0.6
0.4
0.2
0
2 1 0 1 2
y/b
10
8
10
6
10
4
10
2
10
0
E
/
b
u
2w
a
k
e
,
m
a
x
0.9
u
w
a
k
e
/
u
w
a
k
e
,
m
a
x
u
2
1
/
2
w
a
k
e
,
w
2
1
/
2
w
a
k
e
/
u
w
a
k
e
,
m
a
x
u
w
a
k
e
v
w
a
k
e
/
u
2w
a
k
e
,
m
a
x
v
2
1
/
2
w
a
k
e
/
u
w
a
k
e
,
m
a
x
Figure 4. Characteristics of the simulated temporally decaying plane wake for the generation of
inow turbulence: , u
wake, max
= 0.12; , u
wake, max
= 0.10; , u
wake, max
= 0.08; , plane
cylinder wake of Schlichting (1979); , E k
5/3
z law. (a) Mean velocity; (b) streamwise and
spanwise uctuations; (c) wall-normal uctuations; (d) turbulent shear stress; (e) viscous dissipation
rate of turbulence kinetic energy; (f) spanwise spectrum of turbulence kinetic energy.
top boundary condition on boundary layer development should be extremely small.
At the exit of the computational domain, convective boundary conditions were used.
Mass ux at the inow plane was made constant in time by rescaling the velocities
obtained from (7), and corrections to the velocities at the exit plane are also made
to ensure global mass conservation. Periodic boundary conditions were applied in the
homogeneous, spanwise z-direction; u = 0 was applied on the wall.
Boundary layer transition induced by passing wakes 119
(c)
2.0
1.5
1.0
0.5
0 0.02 0.04 0.06 0.08
y
1
/
2
,
+
(d)
2.0
1.5
1.0
0.5
0 0.02 0.04 0.06 0.08
y
(a)
0.8
0.6
0.4
0.2
0
10
1
(b)
3
2
1
0 0.02 0.04 0.06 0.08
1
/
2
,
+
1
/
2
,
+
10
2
10
3
10
4
10
5
1.0
u
Figure 5. Resolution check: symbols, baseline case with x
+
x =3
= 24 (x
+
x =1
= 18.3) and z
+
x =3
= 11
(z
+
x =1
= 8.4):
e
, during transition x = 1.0; , after transition x = 3.0; , spanwise resolution
rened by 50%; , streamwise resolution coarsened by 50%.
2.4. Numerical method
The numerical scheme for the DNS is a parallelized version (by Charles D. Pierce at
Stanford) of the method used by Akselvoll & Moin (1996) and Pierce & Moin (1998).
Convection and diusion terms that involve only derivatives in the wall-normal
direction are treated implicitly, whereas all other terms are treated explicitly. All
spatial derivatives are approximated with a second-order central dierence scheme.
A third-order RungeKutta scheme (Spalart, Moser & Rogers 1991) is used for
terms treated explicitly and a second-order CrankNicolson scheme is used for terms
treated implicitly. The fractional step method is used to remove the implicit pressure
dependence in the momentum equations. Further details can be found in Akselvoll
& Moin (1996). For parallel computation the computational domain is decomposed
in two directions whereas a third direction is complete. When solving the Poisson
equation using fast transforms, a transpose is necessary to switch the un-decomposed
direction. Scalable parallelization is achieved using message passing interface (MPI)
libraries.
2.5. Computational details and resolution check
The governing equations are solved on a rectangular staggered grid. The grid spacings
are uniform in the streamwise and spanwise directions.
Simulation results to be presented in the following sections were obtained on a
(1024, 400, 128) grid in the streamwise, wall-normal and spanwise directions, respec-
120 X. Wu, R. G. Jacobs, J. C. R. Hunt and P. A. Durbin
(c) t/= 32.9
0.6
0.4
0 0.5 1.0 1.5
x
0.2
y
2.0 2.5 3.0 3.5
wake 3
wake 2 wake 1
(b) t/= 32.7
0.6
0.4
0 0.5 1.0 1.5
0.2
y
2.0 2.5 3.0 3.5
wake 3
wake 2 wake 1
(a) t/= 32.5
0.6
0.4
0 0.5 1.0 1.5
0.2
y
2.0 2.5 3.0 3.5
wake 3
wake 2
wake 1
Figure 6. Contours of u over one (x, y)-plane.
tively. The total of 52.4 million grid points is one of the largest that has ever been
reported: compare to 11.0 million, 6.2 million and 17.3 million used by Spalart (1988),
Yang, Spalart & Ferziger (1992) and Rai & Moin (1993), respectively.
In terms of viscous wall units based on the time-averaged local friction velocity
after transition, x
+
x =3
= 24 and z
+
x =3
= 11. When measured using a friction velocity
during transition, x
+
x =1
= 18.3 and z
+
x =1
= 8.4. At the exit, there are 16 points
distributed along the wall-normal direction below y
+
= 9, and a total of 191 points
below y = . The resolution used in Spalart (1988) was x
+
20_1, z
+
6.7_0.34,
with 10 points within 9 wall units.
In order to check the adequacy of the streamwise and spanwise grid resolution,
two complete additional simulations were performed. It was dicult to use a grid size
larger than (1024, 400, 128) due to memory constraints of the computer. Therefore, in
the rst additional simulation the spanwise dimension of the computational domain
was reduced from 0.2L to 0.15L, which is equivalent to 30 at the inlet, 3.94 in the
middle of the plate and 2.18 at the exit. Even such a reduced spanwise dimension
is still suciently wide and we therefore assume most of the dierences, if any,
between the two sets of results are due to the spanwise grid resolution change from
z
+
x =1
= 8.4 to z
+
x =1
= 6.3. In the second additional simulation, the number of grid
points in the streamwise direction was reduced by 50% from 1024 to 768. Except for
these changes, all the other parameters were kept the same as the baseline simulation.
Results from the resolution check are presented in gure 5. Figure 5(a) compares the
three sets of mean velocity proles at two streamwise stations: the rst, at x = 1.0,
is in the transitional region; and the second, at x = 3.0, is in the turbulent region.
Boundary layer transition induced by passing wakes 121
(c) t/= 32.9
0.6
0.4
0 0.5 1.0 1.5
x
0.2
y
2.0 2.5 3.0 3.5
wake 3
wake 2 wake 1
(b) t/= 32.7
0.6
0.4
0 0.5 1.0 1.5
0.2
y
2.0 2.5 3.0 3.5
wake 3
wake 2 wake 1
(a) t/= 32.5
0.6
0.4
0 0.5 1.0 1.5
0.2
y
2.0 2.5 3.0 3.5
wake 3
wake 2
wake 1
Figure 7. Contours of v over one (x, y)-plane. In this and subsequent similar gures, negative
values are contoured by solid lines; positive values are contoured by dashed lines.
The curves show good numerical resolution. Figures 5(b), 5(c) and 5(d) compare
the streamwise, wall-normal and spanwise r.m.s. turbulent intensities, respectively.
Dierences among the three simulations are small. Additional resolution checks can
be found in , 5. In addition to these resolution checks, we also build condence on our
simulation through extensive comparison, presented in , 5, with well-accepted DNS
and experimental data. Comparison with the resolution used in previous channel ow
turbulent spot simulations (e.g. Henningson & Kim 1991) gives further condence that
the present resolution is adequate. Years of turbulence simulation research at Stanford
University has shown that, despite the slow convergence rate with grid renement,
second-order central dierencing has several attractive features. It is energy conserving
and does not carry inherent numerical diusion, as do many high-order upwind biased
schemes.
The time step was xed to be dt = 10
3
=0.00167L/U
ref
, which is equivalent to
0.59 /u
2
, x =3
. Initial velocities were set to the laminar Blasius prole. The ow was
then allowed to evolve for 20 wake passing periods (20 000 dt), and statistics were
then collected for another 20 wake passing periods. Phase averaging was performed
by dividing each pass period into 50 equal subdivisions. The computation was carried
out on the scalable parallel Cray T3E at the Pittsburgh Supercomputing Center, using
up to 512 processors.
3. Visualization of a matured turbulent spot
At the beginning of each period t
m
n
= 1120. It is seen from the gure that upstream of x 0.75 the simulations
follow the Blasius solution quite nicely with only very small deviations. The minor
dierences are due to the impact of the wake on the at plate. Onset of transition
starts at about x 0.7 and by x 2.0 the shape factor has dropped from 2.59
to 1.45. Further downstream the shape factor remains nearly the same, decreasing
only slightly from 1.45 to 1.42 at the exit. Coles (1956) correlation shows that in a
turbulent boundary layer the shape factor drops from 1.48 at Re
= 600 to 1.44 at
Re
= 1100.
Quantitative time-averaged skin-friction data are not available in most experiments
132 X. Wu, R. G. Jacobs, J. C. R. Hunt and P. A. Durbin
(c) t/= 33.1
0 0.25
x
z
0.50 0.75
0.20
0.15
0.10
0.05
(b) t/= 32.9
0 0.25
z
0.50 0.75
0.20
0.15
0.10
0.05
(a) t/= 32.8
0 0.25
z
0.50 0.75
0.20
0.15
0.10
0.05
Figure 17 (ac). For caption see facing page.
on wake-induced transition. Nevertheless, wall shear stress information is crucial since
it provides an important velocity scale for boundary layer theory as well as a necessary
quantity for engineering drag estimation. Figure 21 shows C
f
. Similar to the previous
gure, the Blasius solution and the results from the two simulations are presented.
Within 0.1 6 x 6 0.5, proles of the computed skin-friction coecient follow the
Blasius solution with only a minor over-shoot near the inlet because of the impact of
the wake on the at-plate. C
f
starts to rise beyond x 0.7 and attains a maximum
at approximately 2.15. The streamwise location of the maximum skin friction may
be used as a convenient, well-dened, indicator for the completion of the transition
process. As will be shown next, time-averaged mean streamwise velocity and Reynolds
shear stresses attain their corresponding fully turbulent proles at approximately the
same streamwise station. At the exit, x = 3.5, the computed skin-friction coecient C
f
is 0.00479, 10% higher than that given by Coles correlation. Free-stream turbulence
uctuations, such as those carried by the passing wake in the present case, tend to
increase skin-friction (Hancock & Bradshaw 1989).
Time-averaged streamwise velocities at seven streamwise stations are shown in
gure 22. Figure 22(a) plots u
+
in inner coordinates. The three proles upstream of x =
Boundary layer transition induced by passing wakes 133
( f ) t/= 33.5
0.50 0.75
x
z
1.00 1.25
0.20
0.15
0.10
0.05
(e) t/= 33.4
0.50 0.75
z
1.00 1.25
0.20
0.15
0.10
0.05
(d) t/= 33.3
0.50 0.75
z
1.00 1.25
0.20
0.15
0.10
0.05
Figure 17. Visualization of the evolution of pus using u
/
over the (x, z)-plane of
y = 7.38 10
4
(y
+
= 5.4 at x = 1.75); 0.02U
ref
6 u
/
6 0.14U
ref
with increment 0.02.
1.5 display large deviations from the standard logarithmic prole u
+
= 2.44 ln y
+
+5.0,
though the degree of deviation decreases along the streamwise direction. At x = 2.0
the prole of u
+
still does not possess a well-dened logarithmic slope, indicating that
on average transition is not complete at this station. Further downstream, the three
proles at x = 2.5, 3.0 and the exit (3.5) nearly collapse within 0 6 y
+
6 300. In the
viscous region they follow the law of the wall u
+
= y
+
. In the logarithmic region the
slopes of these proles are in excellent agreement with that of the log law, i.e. 1/
with = 0.41. The intercept of the proles is lower than the standard value 5.0 by
approximately 0.8. This might be attributed to the higher time-averaged skin-friction
value discussed in gure 21. The logarithmic velocity proles produced by the present
simulation are clearly dened. Interestingly, the wake component (Coles 1956) in
the outer part of the boundary layer is also well-dened even though Hancock &
Bradshaw (1989) showed this tends to be suppressed when the intensity of free-stream
turbulence exceeds the friction velocity u
, 10
2
; , 10
2
; .,
/; , 10
2
Re
.
0.5 1.5
x
C
f
2.5
0.006
0.005
0.002
0.001
0 1.0 2.0 3.0 3.5
0.004
0.003
Figure 21. Time-averaged mean skin-friction coecient: , Blasius solution without wake;
, simulation with z
+
x =3
= 11 (z
+
x =1
= 8.4); , simulation with z
+
x =3
= 8.25 (z
+
x =1
= 6.3).
shear stress are in very good agreement with Spalarts DNS. As expected, the free-
stream intensities are higher than Spalart (1988) because of the passing wake. From
x = 2.5 on downstream, changes in the proles are minimal. At the exit (gure 23f)
the maximum value of (u
/
2
)
1/2,+
and its location are in excellent agreement with
Spalart (1988). The prole also develops a shoulder near 0.15 which is commonly
found in low Reynolds number boundary layer ows.
Boundary layer transition induced by passing wakes 137
0.04
y
0.06
1.0
0.4
0.2
0 0.02
0.8
0.6
(c)
u
1.0
y/
1.2
1.0
0.4
0.2
0
0.6
0.8
0.6
(b)
u
10
y
+
100
30
10
0
1
20
(a)
1000
u
+
= y
+
or
u
+
= 2.44 ln y
+
+ 5.0
0.8 0.4 0.2 0
Webster et al. (1996)
u
+
Figure 22. Time-averaged mean streamwise velocity: , x = 0.5; , x = 1.0;
, x = 1.5; - , x = 2.0; , x = 2.5; ., x = 3.0;
e
, x = 3.5 (exit).
138 X. Wu, R. G. Jacobs, J. C. R. Hunt and P. A. Durbin
(e) x = 2.5
3
2
1
0
1
0 0.4 0.8 1.2 1.6
y/
R
e
y
n
o
l
d
s
s
t
r
e
s
s
e
s
( f ) x = 3.5 (exit)
3
2
1
0
1
0 0.4 0.8 1.2 1.6
y/
(c) x =1.5
3
2
1
0
1
0 0.4 0.8 1.2 1.6
R
e
y
n
o
l
d
s
s
t
r
e
s
s
e
s
(d) x = 2.0
3
2
1
0
1
0 0.4 0.8 1.2 1.6
(a) x = 0.5
3
2
1
0
1
0 0.4 0.8 1.2 1.6
R
e
y
n
o
l
d
s
s
t
r
e
s
s
e
s
(b) x =1.0
3
2
1
0
1
0 0.4 0.8 1.2 1.6
Figure 23. Time-averaged Reynolds stresses in outer coordinates; , (u
/
2
)
1/2,+
; , (v
/
2
)
1/2,+
;
, (w
/
2
)
1/2,+
; , (u
/
v
/
)
+
; symbols: at-plate boundary layer of Spalart (1988) at Re
= 1410.
Proles of the time-averaged and normalized (on wall parameters) turbulence
kinetic energy production rate
+
= (u
/
v
/
)
+
u
+
/y
+
at eight streamwise stations
are compared to the DNS of Spalart (1988) and the experimental data of Kim, Kline
& Reynolds (1968) in gure 24. Spalart noticed that his DNS proles of
+
at three
dierent momentum thickness Reynolds numbers are self-similar and also agree very
well with Kim et al. He attributed this to the fact that at such relatively low Reynolds
numbers the decrease of Reynolds shear stress and the increase of mean velocity
gradient cancel each other in the product to a remarkable degree. It is clear from
gure 24 that the present computation faithfully reproduces this feature, as evident
in the self-similarity of the proles at x = 2.5, 3.0 and 3.5 and the agreement with
Spalart (1988) and Kim et al. (1968).
Boundary layer transition induced by passing wakes 139
0.1
0 10 20 30 40
y
+
50 60 70 80
0.2
0.3
0.4
+
c
u
+
/
c
y
+
Figure 24. Time-averaged non-dimensional turbulence kinetic energy production near the wall:
,
Spalart (1988);
e
, Kim, Kline & Reynolds (1968); , x = 0.5; O, x = 0.75; , x = 1.0; ,
x = 1.5; - , x = 2.0; , x = 2.5; ., x = 3.0; , x = 3.5 (exit).
In the transition region the peak of
+
is located away from the wall and there is
also a large over-shoot. This is consistent with the turbulence intensity and Reynolds
shear stress proles presented in gure 23. As the ow approaches the end of transition,
the peak shifts towards the wall and the maximum value of
+
drops. Again, note
that the proles are normalized by the local friction velocity. A clearer picture of
the streamwise evolution of the absolute peak values is given in gure 25. This
gure shows local maxima of time-averaged turbulence kinetic energy, wall-normal
uctuations, Reynolds shear stress and turbulence kinetic energy production. A large
over-shoot in turbulence kinetic energy is seen prior to the completion of transition.
The peak is located at x 1.85. The time-averaged wall-normal uctuations do
not show as noticeable an over-shoot as the turbulence kinetic energy. Production
of turbulence kinetic energy and Reynolds shear stress peak at the same location,
though the degree of over-shoot is much stronger for the former. The location where
the Reynolds shear stress attains its maximum value is the same as the time-averaged
mean skin friction, i.e. x 2.15. In the time-averaged sense this also corresponds
to the end of transition as indicated by the results shown in gures 22, 23 and
24.
Overall, the results presented in this section demonstrate that the present simula-
tions yield correct time-averaged ow statistical properties in the laminar regime
near the inlet and the fully turbulent regime near the exit. Between these two
ends time-averaged mean and second-order statistics proles across the boundary
layer seldom vary monotonically with the increase of x. Completion of transition,
in a time-averaged sense, can be dened as where the mean skin friction reaches
its peak value. At this location (Re
1
0
2
m
a
x
,
c
u
/
c
y
|
m
a
x
m
a
x
,
m
a
x
(b)
0 1 2 3
0.005
0.010
0.015
(a)
Figure 25. Streamwise evolution of maximum Reynolds stresses: - , turbulence kinetic energy;
, wall-normal uctuation; , shear stress; , production of turbulence kinetic energy.
of Liu & Rodi (1991), the location of transition was considered to be where the
turbulent strips merge, and where the turbulence intensity reaches the level pre-
vailing in turbulent boundary layers. The location of merging is dicult to dene,
since it involves transient behaviour and many other subjective factors. Turbulence
intensities also develop over-shoot characteristics in the process of transition. At
the given passing frequency their estimated transition location is x 1.85. Taking
into account the dierent denitions, there is an overall agreement in the transi-
tion location between the present computation and the experiments of Liu & Rodi
(1991).
Boundary layer transition induced by passing wakes 141
0
0 1 2
x
3
0.002
0.004
0.006
C
f
(b)
0.002
0.5 passing periods
0.7 passing periods
0.9 passing periods
0
0 1 2 3
0.002
0.004
0.006
C
f
(a)
0.002
0 passing periods
0.2 passing periods
0.4 passing periods
Figure 26. Streamwise distributions of phase-averaged skin-friction coecient and free-stream
turbulence kinetic energy; upper curves: (C
f
); lower curves: (K) at y = 0.1.
6. Phase-averaged boundary layer properties
Time averaging is not equivalent to ensemble averaging in this non-stationary ow.
The phase average gives a fuller statistical picture. At t
n
= 0 the wake
starts to enter from the top boundary of the inow plane and descends towards
the at plate. Touch-down occurs slightly before t
n
c
f
,
s
t
Figure 27. Streamwise proles of the parameter indicating boundary layer transitional state: ,
based on r.m.s. periodic skin-friction uctuation; , based on r.m.s. periodic Stanton number
uctuation.
when the wake buets the layer. This is distinct from the larger, sustained rise further
downstream where transition occurs, and the boundary layer becomes fully turbulent.
The location of the rise lags the free-stream wake passage as previously discussed.
After the completion of transition no distinct peaks can be seen in the phase-averaged
r.m.s. skin-friction uctuations even though the turbulent layer is still bueted by the
wake. Over the distance where the rise is observed there is a change in the nature
of the uctuations that occurs at around 0.4 passing periods, when the wake reaches
x 0.7. At that point, turbulent spots begin to appear and the uctuation proles
peak deeper in the boundary layer. The appearance changes from that of a bueted
layer to one with self-sustained turbulence. Similar features have also been reported
by Halstead et al. (1997) in multi-stage compressor/turbine measurements.
Figure 27 shows the streamwise distribution of the r.m.s. periodic skin-friction
uctuation coecient
C
f
dened as
C
f
=
__
(C
f
) C
f
_
2
_
1/2
/C
f
. (8)
This coecient is a good indicator for the boundary layers transitional state. Also
presented in the gure is the streamwise distribution of the r.m.s. periodic Stanton
number uctuation coecient
St
, obtained from a heat transfer simulation in which
the wall was slightly heated (Wu & Durbin 1999a). In a laminar ow
C
f
= 0;
after completion of transition in a fully turbulent ow
C
f
0 if the sample size
is suciently large. The denition of
C
f
is precise and does not involve arbitrarily
chosen threshold values. The spike at the inlet station is due to impact of the passing
wake on the at plate. The small peak quickly decays and reaches a local minimum
at x 0.3. The transitional state of the boundary layer at this location is the lowest
prior to the completion of transition. This is again consistent with the notion that
turbulence directly carried by the wake into the near-wall region decays rapidly.
C
f
reaches its global maximum at x 1.2, indicating that at this station the boundary
Boundary layer transition induced by passing wakes 143
(e)
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8
t
n
1
0
2
0.1
0.2
0.3
0.4
1.0
( f )
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8
t
n
0.1
0.2
0.3
0.4
1.0
(c)
1.6
1.2
0.8
0 0.2 0.4 0.6 0.8
1
0
2
0.4
1.0
(d)
1.6
1.2
0.8
0 0.2 0.4 0.6 0.8
0.4
1.0
(a)
0.12
0.08
0.04
0
0 0.2 0.4 0.6 0.8
u
0.04
0.08
0.12
1.0
(b)
0.12
0.08
0.04
0
0 0.2 0.4 0.6 0.8
0.04
0.08
0.12
1.0
Figure 28. Periodic uctuation of streamwise velocity, phase-averaged turbulence kinetic energy
and Reynolds shear stress:
e
, y = 0.002; , y = 0.02; ., y = 0.04; +, y = 0.1; (a, c, e) x = 1.0;
(b, d, f) x = 1.5.
layer has the strongest transitional state, in the sense that the properties of the
boundary layer are most distant from those in either the laminar or turbulent regime.
Figure 27 also shows that within the transition region the degree of upstream and
downstream asymmetry of
C
f
with respect to its peak location is relatively small.
In the remainder of this section we focus on spacetime characteristics of the phase-
averaged mean velocity and second-order turbulence statistics in the transitional
region. Figure 28(a, b) presents the periodic streamwise velocity uctuation (u) u at
x = 1.0 and 1.5 as a function of phase t
n
C
f
, and decreases towards zero in laminar and fully turbulent regions. The peak in
the sine wave has a phase lag behind the free-stream wake. The variation of (K) at
y = 0.002 shows the statistically averaged eect of turbulent spots. Since at x = 1.0
transition is still in the early stage, (K) and (u
/
v
/
) in gure 28(c, e) return to zero after
the statistically averaged eect of turbulent spots has passed. This is dierent from
the situations shown in gure 28(d, f) where turbulence uctuations do not return
to zero for the whole period because transition is already at its late stage. In the
fully turbulent region (u) u, (K) and (u
/
v
/
) inside the boundary layer approach
their time-averaged mean with little sign of phase dependence. Note that the shear
stress inside the free-stream wake is positive. This is a consequence of the wake
orientation.
In the transitional region, the behaviour of (u) u in the central part of the
boundary layer obtained from y = 0.02 is interesting. Figure 28(b) shows the prole
has two dips within one passing period. These two dips correspond to the two peaks
of (K) at the same wall-normal location shown in gure 28(d). The peak of (K)
between t
n
0.5 and 0.6 is clearly from the free-stream passing wake, and the other
one between t
n
= 0.9; , t
n
= 0.9; - , t
n
= 0.05.
its natural level. Figure 30(a) presents contours of instantaneous streamwise velocity,
u U
ref
over an (x, y)-plane at t/ = 8.0. The velocity contours show no sign
of turbulent spots or of transition. Figure 30(b) presents contours of u
/
over the
(x, z)-plane at y/L = 7.38 10
4
at the same instant. The near-wall disturbances
generated at the inlet evolve toward long-streamwise-wavelength waves, with spanwise
modulation. The wake passage is seen more clearly in the v
/
proles of gure 30(c).
These gures show that the boundary layer lters out short-wave inlet disturbances
(Jacobs & Durbin 1998). They decay rapidly and are not the origin of the small-scale
spots that occur beneath higher free-stream wake turbulence.
7.2. Eect of streamwise pressure gradient on transition
Streamwise pressure gradient exerts an overwhelming inuence on the stability of
a laminar boundary layer. A decrease in pressure in the downstream direction, i.e.
favourable gradient, has a stabilizing eect. In the case of a at plate this causes
the critical Reynolds number to become larger than Re
= 1410. J. Fluid
Mech. 187, 6198.
Spalart, P. R., Moser, R. D. & Rogers, M. M. 1991 Spectral methods for the NavierStokes
equations with one innite and two periodic directions. J. Comput. Phys. 96, 297324.
Walker, G. J. 1993 The role of laminar turbulent transition in gas turbine engines: a discussion.
Trans. ASME: J. Turbomachinery 115, 207217.
Boundary layer transition induced by passing wakes 153
Webster, D. R., DeGraaff, D. B. & Eaton, J. K. 1996 Turbulence characteristics of a boundary
layer over a two-dimensional bump. J. Fluid Mech. 320, 5369.
Westin, K. J. A., Boiko, A. V., Klingmann, B. G. B., Kozlov, V. V. & Alfredsson, P. H. 1994
Experiments in a boundary layer subjected to free-stream turbulence. Part 1. Boundary layer
structure and receptivity. J. Fluid Mech. 281, 193218.
Wu, X. & Durbin, P. A. 1998 Boundary layer transition induced by periodic wakes. Trans.
ASME: J. Turbomachinery. (submitted).
Wu, X. & Durbin, P. A. 1999a Numerical simulation of heat transfer in a transitional boundary
layer with passing wakes. Trans. ASME: J. Heat Transfer (submitted).
Wu, X. & Durbin, P. A. 1999b Numerical experiments and modeling of the transitional and
turbulent boundary layers induced by periodically passing wakes. Rep. Center for Integrated
Turbulence Simulation, Stanford University.
Wu, X. & Squires, K. D. 1997 Large eddy simulation of an equilibrium three-dimensional turbulent
boundary layer. AIAA J. 35, 6774.
Yang, K. S., Spalart, P. R. & Ferziger, J. H. 1992 Numerical studies of natural transition in a
decelerating boundary layer. J. Fluid Mech. 240, 433468.
Zhong, S., Kittichaikarn, C., Hodson, H. P. & Ireland, P. T. 1998 Visualization of turbulent spots
under the inuence of adverse pressure gradients. In Proc. 8th Intl Conf. on Flow Visualization,
Italy.